diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -408,6 +408,7 @@
 def TargetX86 : TargetArch<["x86"]>;
 def TargetAnyX86 : TargetArch<["x86", "x86_64"]>;
 def TargetWebAssembly : TargetArch<["wasm32", "wasm64"]>;
+def TargetNVPTX : TargetArch<["nvptx", "nvptx64"]>;
 def TargetWindows : TargetSpec {
   let OSes = ["Win32"];
 }
@@ -1211,6 +1212,12 @@
 }
 def : MutualExclusions<[CUDAGlobal, CUDAHost]>;
 
+def NVPTXKernel : InheritableAttr, TargetSpecificAttr<TargetNVPTX> {
+  let Spellings = [Clang<"nvptx_kernel">];
+  let Subjects = SubjectList<[Function]>;
+  let Documentation = [Undocumented];
+}
+
 def HIPManaged : InheritableAttr {
   let Spellings = [GNU<"managed">, Declspec<"__managed__">];
   let Subjects = SubjectList<[Var]>;
diff --git a/clang/lib/CodeGen/TargetInfo.cpp b/clang/lib/CodeGen/TargetInfo.cpp
--- a/clang/lib/CodeGen/TargetInfo.cpp
+++ b/clang/lib/CodeGen/TargetInfo.cpp
@@ -7356,6 +7356,11 @@
       }
     }
   }
+
+  // Attach kernel metadata directly if compiling for NVPTX.
+  if (FD->hasAttr<NVPTXKernelAttr>()) {
+    addNVVMMetadata(F, "kernel", 1);
+  }
 }
 
 void NVPTXTargetCodeGenInfo::addNVVMMetadata(llvm::GlobalValue *GV,
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
--- a/clang/lib/Sema/SemaDeclAttr.cpp
+++ b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -4868,7 +4868,10 @@
   if (FD->isInlineSpecified() && !S.getLangOpts().CUDAIsDevice)
     S.Diag(FD->getBeginLoc(), diag::warn_kern_is_inline) << FD;
 
-  D->addAttr(::new (S.Context) CUDAGlobalAttr(S.Context, AL));
+  if (AL.getKind() == ParsedAttr::AT_NVPTXKernel)
+    D->addAttr(::new (S.Context) NVPTXKernelAttr(S.Context, AL));
+  else
+    D->addAttr(::new (S.Context) CUDAGlobalAttr(S.Context, AL));
   // In host compilation the kernel is emitted as a stub function, which is
   // a helper function for launching the kernel. The instructions in the helper
   // function has nothing to do with the source code of the kernel. Do not emit
@@ -8744,6 +8747,7 @@
   case ParsedAttr::AT_CalledOnce:
     handleCalledOnceAttr(S, D, AL);
     break;
+  case ParsedAttr::AT_NVPTXKernel:
   case ParsedAttr::AT_CUDAGlobal:
     handleGlobalAttr(S, D, AL);
     break;
diff --git a/clang/test/CodeGen/nvptx_attributes.c b/clang/test/CodeGen/nvptx_attributes.c
new file mode 100644
--- /dev/null
+++ b/clang/test/CodeGen/nvptx_attributes.c
@@ -0,0 +1,22 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --check-attributes --check-globals
+// RUN: %clang_cc1 -triple nvptx64-nvidia-cuda -target-cpu sm_61 -emit-llvm %s -o - | FileCheck %s
+// CHECK: Function Attrs: noinline nounwind optnone
+// CHECK-LABEL: define {{[^@]+}}@foo
+// CHECK-SAME: (ptr noundef [[RET:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RET_ADDR:%.*]] = alloca ptr, align 8
+// CHECK-NEXT:    store ptr [[RET]], ptr [[RET_ADDR]], align 8
+// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[RET_ADDR]], align 8
+// CHECK-NEXT:    store i32 1, ptr [[TMP0]], align 4
+// CHECK-NEXT:    ret void
+//
+__attribute__((nvptx_kernel)) void foo(int *ret) {
+  *ret = 1;
+}
+//.
+// CHECK: attributes #0 = { noinline nounwind optnone "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="sm_61" "target-features"="+ptx32,+sm_61" }
+//.
+// CHECK: !0 = !{ptr @foo, !"kernel", i32 1}
+// CHECK: !1 = !{i32 1, !"wchar_size", i32 4}
+// CHECK: !2 = !{!"clang version 16.0.0"}
+//.